[R-sig-ME] Singular convergence in lmer

[Ben Bolker]

On 10-12-20 10:09 AM, Marie-Hélène Hachey wrote:
> 
> Hi, Sorry I forgot to put a subjet on my last message. I also should
> have added that my dataset has only 49 lines. I am starting to
> understand by reading messages from R-help that the singular
> convergence is a problem aboutthe number of lines of data not being
> enough to estimate effects of many variables? Thank you for
> confirming this! Marie-Helene

  Yes.  The rule of thumb (see e.g. Frank Harrell's book) is that you
should have at least 10-20 data points per parameter you want to
estimate, and that's a best case scenario for continuously distributed
data -- for example, if your Poisson data are fairly sparse they count
for less.